]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
md: propagate BLK_FEAT_PCI_P2PDMA from member devices to RAID device
authorKiran Kumar Modukuri <kmodukuri@nvidia.com>
Wed, 13 May 2026 18:51:52 +0000 (11:51 -0700)
committerJens Axboe <axboe@kernel.dk>
Tue, 26 May 2026 21:51:48 +0000 (15:51 -0600)
MD RAID does not propagate BLK_FEAT_PCI_P2PDMA from member devices to
the RAID device, preventing peer-to-peer DMA through the RAID layer even
when all underlying devices support it.

Enable BLK_FEAT_PCI_P2PDMA unconditionally in raid0, raid1 and raid10
personalities during queue limits setup.  blk_stack_limits() clears it
automatically if any member device lacks support, consistent with how
BLK_FEAT_NOWAIT and BLK_FEAT_POLL are handled in the block core.

Parity RAID personalities (raid4/5/6) are excluded because they require
CPU access to data pages for parity computation, which is incompatible
with P2P mappings.

Tested with RAID0/1/10 arrays containing multiple NVMe devices with
P2PDMA support, confirming that peer-to-peer transfers work correctly
through the RAID layer.

Tested-by: Pranjal Shrivastava <praan@google.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Kiran Kumar Modukuri <kmodukuri@nvidia.com>
Signed-off-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Yu Kuai <yukuai@fygo.io>
Tested=by: Pranjal Shrivastava <praan@google.com>
Link: https://patch.msgid.link/20260513185153.95552-3-kch@nvidia.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
drivers/md/raid0.c
drivers/md/raid1.c
drivers/md/raid10.c

index 5e38a51e349ad3a7cb944c3c830a5f411e373a40..2cdaf7495d92ebf55b3051d0daac7203a0a79850 100644 (file)
@@ -392,6 +392,7 @@ static int raid0_set_limits(struct mddev *mddev)
        lim.io_opt = lim.io_min * mddev->raid_disks;
        lim.chunk_sectors = mddev->chunk_sectors;
        lim.features |= BLK_FEAT_ATOMIC_WRITES;
+       lim.features |= BLK_FEAT_PCI_P2PDMA;
        err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
        if (err)
                return err;
index 64d970e2ef50fda96a632956c39ffcca52b0fe3e..cc628a1be52c7d53590a8aff3ed72243017e4367 100644 (file)
@@ -3208,6 +3208,7 @@ static int raid1_set_limits(struct mddev *mddev)
        lim.max_hw_wzeroes_unmap_sectors = 0;
        lim.logical_block_size = mddev->logical_block_size;
        lim.features |= BLK_FEAT_ATOMIC_WRITES;
+       lim.features |= BLK_FEAT_PCI_P2PDMA;
        err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
        if (err)
                return err;
index 39085e7dd6d26d540a61e821f888eaf67e29ce8d..f905dc391b74534fbc2f2f2c5506eeb46b1269a5 100644 (file)
@@ -3941,6 +3941,7 @@ static int raid10_set_queue_limits(struct mddev *mddev)
        lim.chunk_sectors = mddev->chunk_sectors;
        lim.io_opt = lim.io_min * raid10_nr_stripes(conf);
        lim.features |= BLK_FEAT_ATOMIC_WRITES;
+       lim.features |= BLK_FEAT_PCI_P2PDMA;
        err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
        if (err)
                return err;